EVM vs PER Plot Not Promising for PSNI

1. EVM vs PER Plot Not Promising for PSNI Brian Hart, David Skellern (Cisco Systems) skellern@cisco.com, brianh@cisco.com TGk July 2003 Hart/Skellern Cisco

2. Background • IEEE 802.11-03/218r2 (file 11-03-218r2-K-PSNI_Measurement_V2.ppt) proposed the use of a measure named PSNI to quantify the quality of the signal that is able to be recovered by a particular receiver • In document IEEE 802.11-03/531r0 we suggested that a new measure was not needed because an appropriate measurement of output signal quality was already available in EVM which: • works for all digital modulations • has the desired properties specified on slides 9 & 10 of doc: IEEE 802.11-03/218r2 • can provide a direct indication of observed S/(N+I) considering all channel impairments and implementation losses when measured at the demodulator • is defined already in the standard as a signal quality measure • can be measured with commercially available test equipment Hart/Skellern Cisco

3. Background cont. • However, discussion in TGk at the May 2003 concluded that • Simulations and/or measurements were needed to verify the validity of EVM as an output signal quality measure • Even if EVM was a valid measure, it was a measure that needed to be extended to take into account the implementation characteristics of the receiver from the demodulator to the output (principally the FEC block) • This submission is about the validity of EVM and reports on an investigation of the relationship between EVM and PER across different channels and practical systems. Hart/Skellern Cisco

4. Investigation of EVM validity • We thought that if any useful standardized PSNI measure existed it would be based on EVM because it captures all the degradations present on a communications link • in proportion to the impact they have • at a point where the cumulative degradation is easy to estimate • includes: quantisation, non-ideal converters, clock frequency offsets and phase noise, carrier frequency offsets and phase noise, power-amplifier distortions, multipath, co-channel interference, adjacent channel interference, other interference, thermal noise, and poor receiver algorithms. • We therefore expected that if we plotted average EVM vs average PER for an IEEE 802.11a link operating at a given rate over a range of practical implementations and typical channels we would find a useful relationship between EVM and PER. Hart/Skellern Cisco

5. Investigation of EVM validity • We generated results using a Matlab 802.11a modeling framework • with a range of channels, from AWGN to Rayleigh, that we believe from our channel measurements to be representative of the indoor office environment • with a range of practical hardware settings and implementations, including ADC resolutions, datapath widths, Viterbi algorithm details, synchronization options, channel tracking methods and AGC strategies “practical hardware settings and implementations” = workable designs, many of which are in current manufacturers’ implementations • Simulation set-up • there are over 250 plotted simulation points • generally the simulations wait for 75 1000-byte packets in error or 750 packets, whichever comes first • a few points (<< 5%) at high PER are based on fewer packets Hart/Skellern Cisco

6. EVM vs PER results for 6 Mbit/s Genie-aided EVM is EVM calculated using the known transmitted constellation point. Hart/Skellern Cisco

7. Conclusion • These results show no useful relationship between EVM and PER across different channels and practical systems. • While further study is needed, we conclude that obtaining an output data quality measure that would be useful across different designs and implementations (manufacturers) is unlikely in the short term – and not in any timeframe commensurate with timely completion of an 11k draft. • We therefore propose that text related to PSNI not be included in the draft. Hart/Skellern Cisco